Dynamic Speech Model Selection for Aviation Transcription
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Automatic Speech Recognition (ASR) systems in aviation domains face challenges with high memory usage and latency due to the need for large vocabularies of unique keywords and acronyms, which are detrimental to recognition accuracy and latency, especially in resource-constrained cockpit environments with limited external cloud support.
Innovation Solution
A transcription system with an event detector and intelligent keyword generator dynamically selects or generates a constrained speech model based on flight phase changes, using a database of pre-built models or creating a new model to cover unique keywords, optimizing memory usage and improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large vocabulary of unique keywords and acronyms is stored in the ASR model to improve transcription accuracy in the aviation domain, then recognition accuracy is improved, but memory usage increases and latency increases
Solution Approach 1:
The patent segments the large aviation vocabulary into multiple constraint files organized by categories (aircraft types, flight phases, airport codes, etc.). Each constraint file contains a subset of keywords relevant to specific contexts, allowing the ASR model to load only necessary portions rather than the entire vocabulary at once.
Solution Approach 2:
The system dynamically loads and switches between different constraint files based on the current flight context, aircraft type, and operational phase. This dynamic approach allows the ASR model to adapt its vocabulary to the specific situation, maintaining high accuracy while minimizing memory usage by only loading relevant keywords when needed.
2Measurement precision
If a large vocabulary of unique keywords and acronyms is stored in the ASR model to improve transcription accuracy, then recognition accuracy is improved, but latency increases
Solution Approach 1:
By segmenting the vocabulary into smaller constraint files, the system reduces the time required to initialize and process the ASR model. Each constraint file can be loaded and processed more quickly than a single large vocabulary, reducing overall latency while maintaining comprehensive coverage through selective loading.
Solution Approach 2:
The dynamic loading mechanism allows the system to load only the necessary constraint files for the current operational context, avoiding the latency penalty of loading the entire vocabulary. The system can quickly switch between different constraint sets as flight conditions change, maintaining responsiveness and low latency.
3Quantity of substance
If the ASR model is constrained to reduce memory usage and latency, then resource efficiency is improved, but transcription accuracy may deteriorate
Solution Approach 1:
The constraint file system is designed to be universal and multi-functional, with constraint files that can be applied across different aircraft types, flight phases, and operational contexts. Each constraint file serves multiple purposes and can be combined with others to cover the full range of aviation terminology, ensuring comprehensive accuracy while maintaining resource efficiency.
Solution Approach 2:
The system dynamically selects and combines multiple constraint files based on the current flight context, ensuring that the most relevant keywords are always available for accurate transcription. This dynamic approach maintains high accuracy by adapting the vocabulary to match the specific operational situation while keeping memory usage low through selective loading.
4Device complexity
If the ASR model is constrained to execute locally on edge devices with limited resources, then device complexity is reduced and local execution is enabled, but the model size must be reduced
Solution Approach 1:
The segmentation of the vocabulary into modular constraint files allows the ASR model to be deployed on resource-constrained edge devices in the cockpit. Each constraint file is a self-contained unit that can be loaded independently, enabling the model to operate with limited memory while still providing comprehensive aviation terminology coverage when all constraint files are combined.
Solution Approach 2:
The dynamic constraint loading system enables the ASR model to adapt its vocabulary size and content based on the available resources and current operational needs. This allows the model to maintain high versatility and coverage by loading additional constraint files when resources permit, while still functioning effectively with reduced model size when executing locally on edge devices with limited resources.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Systems and methods are provided for the selection of a speech model for automatic speech recognition during runtime of a transcription system, the system includes an event detector to determine one of a number of flight events that include flight plan changes and phase transitions based on data received from a set of inputs; an intelligent keyword generator to collate a set of keywords associated with the flight plan information and to generate a wordlist in response to a determination by the event detector of flight plan changes or flight phase transitions; and a processor to determine whether the wordlist is covered by a current speech model implemented in the speech recognition wherein if the wordlist is not covered by the current speech model, then the processor to select a pre-built speech model that covers the wordlist for use as the current speech model in the automatic speech recognition.